Computational Intelligence Modeling of Pharmaceutical Roll Compaction

Datasets ordinarily include a huge number of features (attributes) with irrelevant and redundant features. Selecting a subset of relevant properties from a large set of features that describe a dataset is a challenging machine learning task. Feature selection algorithms explore the data to eliminate noisy, irrelevant, redundant data, and simultaneously optimize the classification performance. In biology, for instance, the advances in the available technologies enable the generation of an enormous number of biomarkers that describe the data. Selecting the more relevant biomarkers along with a very high prediction performance over the data can be a daunting task, particularly if the data are high-dimensional. In this study, we have been developed new variants of the native bio-inspired optimization algorithms such as binary antlion optimization (BALO), chaotic antlion optimization (CALO), Lèvy antlion optimization (LALO), binary grey wolf optimization (BGWO), and much more. All the proposed optimization algorithms were compared to two well-known algorithms used in feature selection, namely particle swarm optimization (PSO) and genetic algorithm (GA). A set of assessment indicators were used to evaluate generalization ability and prediction performance of the algorithms in the classification problem over 21 datasets from the UCI repository (10 datasets in the regression problem). The experimental results prove the capability of the proposed variants of the native optimization algorithms to explore the search space for the optimal feature subset regardless of the initialization methods and the used stochastic operators. In the pharmaceutical industry, to develop new formulations or products and to optimize manufacturing processes are often used the exploitation of knowledge on the causal relationship between product quality and attributes of formulations. With the big data captured in the pharmaceutical product development practice, computational intelligence (CI) models, based on machine learning and bio-inspired optimization algorithms, could potentially be used to identify critical quality attributes (CQA) and critical process parameters (CPP), for the formulations and manufacturing processes. The principal objective of our study in the pharmaceutical field is to evaluate the robustness of machine learning techniques combined with bio-inspired optimization algorithms in modeling tablet manufacturing processes. More precisely, our effort is focused on the prediction of tablet properties such as porosity and tensile strength from powder and ribbons characteristics. For this purpose, roll compaction experiments were performed with various pharmaceutical excipients (MCC PH 101, MCC PH 102, MCC DG, Mannitol Pearlitol 200SD, Lactose, and binary mixtures), leading to datasets with a large number of attributes (features). The modeling efficiency is evaluated regarding the average of selected features size (reduction) and the root mean square error (RMSE). We have remarked that the predicted results were in good agreement with the actual experimental data.